o oi@sddlmZddlmZddlmZmZmZmZddl Z ddl m Z ddl m Z ddlmZd eeefd eeefd efd d ZGdddZGdddZdS))inf)Path)CallableDictOptionalUnionN)Module) AverageMeter) TrainerStateepochmetricreturncCsd|d|dS)z.Generate the filename in the model checkpoint.z model_epoch=z _metricValue=z.pt)r r rrF/home/ubuntu/.local/lib/python3.10/site-packages/kornia/x/callbacks.pydefault_filename_fcnsrc @sXeZdZdZ   ddedededed d f d d Zd e dede ee fd e fddZ d S) EarlyStoppinga+Callback that evaluates whether there is improvement in the loss function. The module track the losses and in case of finish patience sends a termination signal to the trainer. Args: monitor: the name of the value to track. min_delta: the minimum difference between losses to increase the patience counter. patience: the number of times to wait until the trainer does not terminate. max_mode: if true metric will be multiply by -1, turn this flag when increasing metric value is expected for example Accuracy **Usage example:** .. code:: python early_stop = EarlyStopping( monitor="loss", patience=10 ) trainer = ImageClassifierTrainer( callbacks={"on_epoch_end", early_stop} ) Fmonitor min_deltapatiencemax_moderNcCs8||_||_||_||_d|_|rt nt|_d|_dS)NrF)rrrrcounterr best_score early_stop)selfrrrrrrr__init__=s zEarlyStopping.__init__modelr valid_metriccCs||jj}|jr||jkn||jk}|r||_d|_n%|jr'||j|jkn||j|jk}|sA|jd7_|j|jkrAd|_|jrOtd|dt j St j S)Nrr Tz3[INFO] Early-Stopping the training process. Epoch: .) ravgrrrrrrprintr TERMINATETRAINING)rrr rscoreis_bestis_within_deltarrr__call__Ps"  zEarlyStopping.__call__)rrF)__name__ __module__ __qualname____doc__strfloatintboolrrrr r r(rrrrr#s" &rc @sbeZdZdZ  ddededeedefded df d d Zd e d e de ee fd dfddZ dS)ModelCheckpointaGCallback that save the model at the end of every epoch. Args: filepath: the where to save the mode. monitor: the name of the value to track. max_mode: if true metric will be multiply by -1 turn this flag when increasing metric value is expected for example Accuracy **Usage example:** .. code:: python model_checkpoint = ModelCheckpoint( filepath="./outputs", monitor="loss", ) trainer = ImageClassifierTrainer(..., callbacks={"on_checkpoint", model_checkpoint} ) NFfilepathr filename_fcn.rrcCsD||_||_|p t|_|rt nt|_||_t|jjddddS)NT)parentsexist_ok) r2rr _filename_fcnr best_metricrrmkdir)rr2rr3rrrrrs  zModelCheckpoint.__init__rr rcCsZ||jj}|jr||jkn||jk}|r+||_t|j|||}t||dSdS)N) rr!rr7rr2r6torchsave)rrr rvalid_metric_valuer&filenamerrrr(s zModelCheckpoint.__call__)NF)r)r*r+r,r-rrr0rrr/rr r(rrrrr1ks  &r1)mathrpathlibrtypingrrrrr9 kornia.corerkornia.metricsr utilsr r-r/r.rrr1rrrrs     &H